Learning of Perceptual Semantics in Cognitive Robots through Social Interaction

Abstract

There is a growing interest to develop cognitive virtual agents. In virtual environment, these agents are not only resembled to humans but also exhibit human-like behavior. Learning of these agents are quite different than the ways through which humans acquire knowledge [1]. Most agents collect data from users and the environment and then implements efficient algorithms on collected data to infer knowledge [2]. In this methodology, a model is trained and the trainer provides on-demand information to model through learning process. Trained model is used as a perceptual semantic capability of an agent. Machine learning differs from human learning because it requires an active social partner in learning environment.

In virtual environment, an agent is able to interact with environmental objects and gather knowledge to differ between various objects. A trained model is able to simulate previously learned knowledge with current perspective [1]. For this purpose, a model is trained and used with middleware. Graph database is used to store semantics data with a set of training examples to infer cognition for knowledge representation. Whenever a robot will face an object, it will perform detection by perceiving it and recognition by semantic knowledge.